|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
67.60.98.130
In Reply to: RE: Linear Power Supplies for ATX & P4 posted by GStew on May 31, 2010 at 09:17:49
In this thread:
http://www.audioasylum.com/cgi/t.mpl?f=pcaudio&m=75244
Jack Wong wrote:
"I just saw your thread and problem. The following may help:
1. Use a CRC filter to absorb unnecessary voltage. The first cap smaller than the second one, a resistor in between. This will reduce surge current, smooth ripples and reduce dissipation of the regulators. For 5V regulator @ 2.4A, using a LM317K (needs min. 3V to work), say a DC supply of 12V (after rectification), a 1.5 ohm 10W resistor will absorb 3.6V of voltage and 8.6W of heat. The LM317K will see a 8.4V supply and dissipate about 8.2W.
Other methods
2.Increase the size of your heat sink or use separate heat sinks for
different regs. Smaller one for the 12V andf 3.3V.
3. Use low dropout reg. as they can work with a min. of 1V potential difference.
4. Use optimal AC voltage for the reg. to reduce heat."
My response...
On the over-temp condition of my supply... First, I'm using LT1083 fixed-voltage regulators, which are fairly low dropout voltage devices at about 1.5v. In the past I'd heard that more voltage is sonically preferable, but I didn't go hog-wild on this... The 12v regulator is dropping a bit more than 6v, the 5v a bit more than 4v. But one option I'm considering is transformers that give a lower raw supply voltage.
Still, that's a fairly costly solution (about $100 for another pair of transformers), so my next try will be changing the regulator chips from the fixed-output versions of the LT1083's to the variable ones. I used these in the raw supply for my PicoPSU-based hybrid supply and they run heatsink temps of over 140F in that setup with no issue... The fixed-output ones are going into protection at about 130F, so this may solve the issue by itself. Also, I'll change the insulators from Silpad-types to mica which has a lower thermal resistance.
If that doesn't do it, next is to move the low-current regs (3.3v and ATX 12v) to their own heatsinks.
And it that doesn't do it, I've been slowly working on a new case layout that will give me more room for heatsinks. If I think I can finish that soon, then I'll go that direction. If not, I'll look at getting lower-voltage transformers to reduce the voltage drop.
The CRC filter is a good thought and one I hadn't considered... But it would be hard to implement in my current packaging and still be able to use it in the Zalman case.
Thanks!
Greg in Mississippi
Everything matters!
Follow Ups:
Jack,
Not 100% sure what the issue was and not 100% sure I've solved it, but it's running longer now than it has before.
I suspect the fixed-voltage LT1083's were having momentary cutouts. Not sure why... they were well within their temp and current ranges. If it keeps working ok, I won't spend the time to try and determine the ultimate reason... just not worth spending the time when I have a lot of other things I need to / want to do.
I went back and looked at the datasheets for both the fixed and variable 1083's and realized I'd made a units mistake... they list max temps in the 150-200C range, I was reading temps of 130F! Big difference here... none of the chips should have had issues with this. It's hot... hotter than I like, but not over-temp hot.
In any case, I changed out the 5v ATX and 12v P4 regs from the fixed to the variable last night... it went longer before it cut out, so I thought I might have a bad chip in at the 12v ATX position, which was my last fixed one. So I replaced it with the 12v LT1083 that had been in as the 12v P4 chip and again it went longer... long enough that I thought it was fixed, but after putting the cover back on, it cut out again. So I took off the cover, did what I was trying to avoid and drilled ventilation holes above the heatsink, and tried it again. When that didn't fix it, I replaced that chip with another LT1083CP and for the last couple of hours, it's been good.
Preliminary sonic diffs with the fully linear supply... Significant improvement over the linear-hybrid supply (modified PicoPSU)... not as large as the difference between a modified Antec with additional filtering and the linear-hybrid, but VERY nice and worthwhile. Greater silence between sounds, more definition across the board, instrument and vocal textures are more audible, greater coherence between fundimentals and overtones. Stage width is surprisingly wider, didn't expect that. Curious how breakin will affect it... with some Black Gates & Jensen caps, it will be at least a couple of weeks before it finally settles in.
One interesting thing to note is that the motherboard is now stable with an even slightly lower voltage... I'm down to .72500 now with no sign of glitchiness at 192 oversampling, was at .73750 before with some glitchiness. This might be slight differences in voltage levels... need to measure exactly what my PicoPSU was putting out and what this one does. But I suspect it is having lower noise on the supply. Also, my Q&D check of listening for the level of interference on a battery-powered FM radio held near the computer shows it being much quieter than before.
Drawbacks are the time it took to debug it and get it working right... and the more-involved startup process. But I generally leave my cMP on full-time except when working on it or when thunderstorms are predicted, so that won't be a big issue.
While not quite as hard-core as your fully battery-powered setup, it is still pretty trick, with fully separate highly-over-spec'd supplies for each voltage. I want to play around with with different regulators along with powering the P4 from my dirty AC line (easy to do, just use my old supply for the P4).
I've been thinking about how to do a how-to guide... will post more about that in the next few days. Many thoughts here... there's not an easy way to do this and make it totally universal.
And I'll post more pix soon... my wife has the camera while she's off visiting her mom (in Hawaii!) and my cell-phone camera pix are just not that good.
Again, thanks for all your help.
Greg in Mississippi
Everything matters!
Greg, You are welcome. Great to hear you've fixed the problem.
It ran ok all night, so changing out all the fixed-voltage LT1083's to the variable ones seems to have fixed the issue.
Everything matters!
Resolving the problem... FINALLY!
I was wrong. When I last posted, my issues with my linear supply were not fixed. The computer would always turn off after running awhile... It might cut out in as little as 30 minutes, it might run as long as 8-10 hours, but it would ALWAYS shut down at some point. Sometimes it would reboot... Mostly it would just turn off.
At first, I focused on the possibility of the regulators overheating and going into thermal protection shutdown (unlikely as the chips are rated for 125C and were only getting to 60C).
I tried the following:
1. Installed separate small heatsinks for the ATX 12v and 3.3v regulators to reduce their temps.
2. Changed the regulator biasing scheme. I configured my regulators based on this thread by John Bau, the designer of Spica speakers, where he worked to optimize the performance of 317/337 type regulators: http://www.diyaudio.com/forums/power-supplies/143539-another-look-lm317-lm337-regulators-5.html#post1891710 . When I setup my linear-hybrid power supply last year, I used the optimization scheme he published in the first few posts of the thread. When I configured this fully linear supply, I used his final optimization scheme, which seemed to set a higher current level for the internal amplifier of the regulator chip, possibly promoting thermal protection shutdown. So I went back to the earlier scheme.
3. Changed out all the regulator chips to ensure I didn't have any defective ones.
4. Change out all the rectifier diodes (a longshot, as failing rectifiers just fail in my experience, they don't go intermittent).
None of these prevented the eventual shutdown.
I then focused on the controlling voltages. When I first configured the supply, I found that the motherboard would run ok after booting if you turned off the 5vSB, PWR_OK, and -12 lines, so I used one 9v battery regulated down to 5v for the 5vSB & PWR_OK lines and two 9v batteries in series regulated down to provide the -12v. And after the motherboard was running, I turned them off to prevent running the batteries down. I later tried a single un-regulated 9v battery (Thanks Gene!) for the -12v and that worked just as well as the regulated -12v, so I left it that way. But all the while using this control voltages scheme, it still turned off eventually.
After playing around with this, I found that the 5vSB and PWR_OK being off were causing the turn-offs. As long as these voltages were provided, it would not turn-off. But cut them off and the board would run at first, but eventually shut down. So I've installed an additional small linear supply to provide 5v for the PWR_OK and 5vSB. I now only turn off the -12v (still provided by a 9v battery) after the motherboard is running.
Now it stays on all the time!
Sonics?
It is still in the midst of component break-in (especially with the Black Gates), but initial indications are that it is a nice upgrade from the Linear/Hybrid supply with a modified PicoPSU, most noticably with greater definition at both frequency extremes, greater extension in the highs, and an increase in apparent dynamics. Interestingly, it is not nearly as large of an upgrade over the Linear/Hybrid as that was over the stock or modified Antec. Based on this, I'd still recommend the Linear/Hybrid as a viable alterative for those who want a very significant sound quality improvement with less work, cost, and hassle than the fully linear supply. But if you want the ultimate, the fully linear supply is the entry point.
Next stages?
Of course I can't leave well-enough alone. I've already gone back to the final regulator configuration from John Bau's thread and changed out the voltage-setting resistor bypass capacitors from Elna Silmec to Black Gate standards. (BTW, there is an interesting aside here... The diffences in sound quality of the Silmecs the Black Gates came through very clearly in my system, even though this is a non-critical component position in the power supply of the computer, not the sound card, DAC, or ancilliaries! I personally prefer the Black Gates with an overall greater sense of flow, a more palpable midrange, and ultimately a smoother and more musical sound, but many would prefer the Silmecs with a more dynamic sounding bass and a bit more 'technocolor' outlining of the sound. I do need to try the Nichicon Muse and Fine Gold caps here along with Oscons, Mundorfs, Sikorels, and Jensens as the Black Gate supply is dwindling fast! But my question for those who understand this better than I is why can component selections be clearly heard even though this is not in the 'audio' path?).
In the search for further sonic upgrades and operational improvements, next things to try are:
1. Feed the 5v for the 5vSB & PWR_OK from the main 5v supply instead of a separate supply. Simplified grounding and reducing the transformer and rectifier count may make this worthwhile.
2. Various regulator upgrades. I have the Linear Tech LT3080's here to try (Thanks for pointing out the comment in the K&K Audio DAC info, Rick!). They are limited to 1.1A which will work for the ATX 3.3v/12v and P4 12v, but not for the ATX 5v... But the chips can be paralleled for greater current handling and I'll use four to supply the 2.2A for the ATX 5v, one connected to each pin on the ATX connector.
Beyond these monolithic regulators, I'd like to try the AMB Sigma 11 discreet regulators (http://www.amb.org/audio/sigma11/), the high-current version of the Salas Simple Shunt (http://www.diyaudio.com/forums/power-supplies/143693-simplistic-salas-low-voltage-shunt-regulator.html and http://www.diyaudio.com/forums/power-supplies/168631-5v-high-current-low-noise-regulator.html, and of course Paul Hynes highly-regarded regs.
3. Low-noise raw DC supplies ala' John Swenson's design published elsewhere in this forum. While I used good-quality soft-recovery diodes and filtering caps, I suspect John's choke-filtered design will do a better job of minimizing supply-produced noise. But implementing these will require re-packaging and my supply will then no longer fit in the Zalman case, so it's a bit down the road.
4. Powering the P4 supply from my 'dirty' AC circuit. This is easy, I can use the old Linear/Hybrid to power the P4 and keep the fully linear ATX powering the rest of the computer. I'm very curious what effect this will have.
5. Getting the supply to work with the Dual-Bios motherboards such as the newly recommended GA-H55M-UD2H. I've been unable to get this simple (single switch to activate PWR_OK, no relays, no PicoPSU) linear supply to work with the GA-G31M-UD2H board and figure it won't work with any of the newer dual-bios boards. JackWong, have you tried the simple scheme with one of these boards with any success?
I expect that using a PicoPSU connected to the 5vSB, PWR_OK, and PS_ON# connections and controlling a relay that turns on the AC connections for the linear ATX 3.3v, 5v, 12v, and P4 12v will provide the sequencing needed for the dual-bios boards. My gut feel is that it MAY take a slight delay on the PWR_OK signal (to allow the linear supplies to power-up), but that's a simple circuit to implement (555 timer and a relay on the PWR_OK line). I'll give that a try soon.
6. Batteries (someday!).
Enough for now, I'll post again soon with my thoughts on how-to's.
Greg in Mississippi
P.S. Current usage is getting down there due to the slightly lower voltage, with the ATX 5v @ 2.2A, 3.3v @ .3A, 12v @ .1A, and the P4 at .5A, giving just over 19w!
Everything matters!
I had a good thing happen on my cMP this afternoon!
I had not been totally happy with the sound after the last set of upgrades in the fully-linear supplies... while the extended and more detailed bass and treble were evident, the tonality was not right and it wasn't 'tuneful'. I first thought it was component break-in... the that it might be the Black Gates I put in, which were not a popular size (330uf/16v) and therefore might have been older stock... either needing longer than normal to break in and sound good or maybe just not as good of a capacitor as the others I've used.
So I planned to push forward on trying the LT3080 regs this weekend, which in addition to possibly being better than the LT1083's that I'm currently using, also replaces those caps with smaller film caps.
I had planned the following sequence of mods:
1st, rebuild my linear 'dirty' supplies (USB, screen, HDD) onto a smaller heatsink so I can use the 'overkill' heatsink on that supply to build up the new LT3080 regulator sections and make the switchover on the fully-linear supply easier and quicker. Also while I was doing this, I'd replace the fixed output (5v, 12v) LT1083's with some of the variable output ones I pulled out of the fully-linear supply while I was troubleshooting... I saw a comment from either Thorsten or FMAK this week suggesting the variable versions of regulators like this are better because you can use a bypass cap on the voltage-setting resistor, lowering the output noise level. On the fixed ones, this resistor is inside the chip... on the variable ones, you add this resistor to ground outside the chip. Made perfect sense and sounded well-worth implementing.
2nd, reassembly my hybrid-linear supply (I stole the LT1083 regulators out of this when I was troubleshooting the fully-linear supply) so I can use it's voltages to test the LT3080 regulator assembly as I build it up AND put it into the system so I can listen while I work.
3rd, build-up and test the LT3080 regulator assembly.
4th, install the LT3080 regulator assembly in the fully-linear supply.
But during the 1st step, I believe I found the issue causing the disappointing sonics... the solder connection of my power cord to the fuse-holder had broken loose and while it touched and still worked, it was not secure. So I fixed this while doing the dirty supply rebuild.
When I put it back in, the tonality was back to what I remembered (with the added bass and treble extension) and the tunefulness was back!
YAH!!!!
Now I can put the LT3080 upgrade to the side and work on DAC upgrades that I've been putting off while troubleshooting the fully-linear, especially getting a Buffalo-II DAC up and running!
And those Black Gates apparently are ok!
Listening to Dire Straits' Making Moves right now... The opening 15 seconds blew me away, then skipped to track three and WOAH!
Last time I listened to this, sometime last summer, probably before the separately-powered Juli@, it sounded muddled and not too dynamic. Tonight it is SEX! Percussion is exciting, it does not muddle up even when very complex (and very overdubbed), very intelligible with a very strong sense of musical flow and musical sense. Dang, this is what we do this sh*t for!
Before, I was listening to my favorite recordings and half-enjoying them.
Now, I am seeking out recordings of note and LOVING them. I'm not hearing much that I hadn't heard before (some small details), but the presentation is very inviting and level of musicality above any previous configuration... I WANT to listen to everything in my collection!
Gotta love when it works right!
A few thoughts from this:
1. This impairment and subsequent resolution involved my 'dirty' supplies. I still don't know why, but these seem almost as important as the 'key' motherboard, sound card, and DAC supplies. I'm not only using linear supplies here, but have found that using super-high quality parts is also important.
2. I'm finally hearing the full performance of the fully-linear supply on my cMP (while component breakin is not complete, it is more than 3/4 through).
Now I would say that it provides ALMOST as much improvement over the hybrid-linear as that supply provided over a modified Antec. Well worthwhile!
3. I'm still considering an extended how-too, but think that it will be hard to do too specific of an article due to the packaging. They need to be considered in context of a specific computer case implementation. My current supply just barely fits in my Zalman along with my linear 'dirty' supplies and DAC supplies. My next generation with choke-filtered supplies will take up more than 50% more volume!
And something like JackWong96's battery supplies take up even more volume!
I'm thinking of a how-to that covers several levels of implementation:
1. Basic using commercially-available linear supplies and regulators.
2. Mid-level using built-up linear supplies and higher-quality regulators.
3. Advanced using high-test parts and construction techniques.
4. Super-advanced using batteries.
All will require some decisions by the builder on packaging for their implementation.
Anyone interested in being the guinea pig for the basic supplies?
Greg in Mississippi
Everything matters!
Anyone interested in being the guinea pig for the basic supplies?
Depending on your time frame, count me in!
No one here remembers the bending of our minds
Post a Followup:
FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
This post is made possible by the generous support of people like you and our sponsors: